fn log1p_64(x: f64) f64x: f64103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178fn log1p_64(x: f64) f64 {
const ln2_hi: f64 = 6.93147180369123816490e-01;
const ln2_lo: f64 = 1.90821492927058770002e-10;
const Lg1: f64 = 6.666666666666735130e-01;
const Lg2: f64 = 3.999999999940941908e-01;
const Lg3: f64 = 2.857142874366239149e-01;
const Lg4: f64 = 2.222219843214978396e-01;
const Lg5: f64 = 1.818357216161805012e-01;
const Lg6: f64 = 1.531383769920937332e-01;
const Lg7: f64 = 1.479819860511658591e-01;
const ix: u64 = @bitCast(x);
const hx: u32 = @intCast(ix >> 32);
var k: i32 = 1;
var c: f64 = undefined;
var f: f64 = undefined;
// 1 + x < sqrt(2)
if (hx < 0x3FDA827A or hx >> 31 != 0) {
// x = -1.0
if (ix == 0xBFF0000000000000)
// log1p(-1) = -inf
return x / 0.0;
// x < -1.0
if (hx >= 0xBFF00000)
// log1p(x < -1) = nan
return (x - x) / 0.0;
// |x| < 2^(-53)
if ((hx << 1) < (0x3CA00000 << 1)) {
if ((hx & 0x7FF00000) == 0) {
mem.doNotOptimizeAway(@as(f32, @floatCast(x)));
}
return x;
}
// sqrt(2) / 2- <= 1 + x < sqrt(2)+
if (hx <= 0xBFD2BEC4) {
k = 0;
c = 0;
f = x;
}
} else if (hx >= 0x7FF00000) {
return x;
}
if (k != 0) {
const uf = 1 + x;
const hu = @as(u64, @bitCast(uf));
var iu = @as(u32, @intCast(hu >> 32));
iu += 0x3FF00000 - 0x3FE6A09E;
k = @as(i32, @intCast(iu >> 20)) - 0x3FF;
// correction to avoid underflow in c / u
if (k < 54) {
c = if (k >= 2) 1 - (uf - x) else x - (uf - 1);
c /= uf;
} else {
c = 0;
}
// u into [sqrt(2)/2, sqrt(2)]
iu = (iu & 0x000FFFFF) + 0x3FE6A09E;
const iq = (@as(u64, iu) << 32) | (hu & 0xFFFFFFFF);
f = @as(f64, @bitCast(iq)) - 1;
}
const hfsq = 0.5 * f * f;
const s = f / (2.0 + f);
const z = s * s;
const w = z * z;
const t1 = w * (Lg2 + w * (Lg4 + w * Lg6));
const t2 = z * (Lg1 + w * (Lg3 + w * (Lg5 + w * Lg7)));
const R = t2 + t1;
const dk = @as(f64, @floatFromInt(k));
return s * (hfsq + R) + (dk * ln2_lo + c) - hfsq + f + dk * ln2_hi;
}